LATEST: A learning-based automated thermal environment control system

Abstract

Thermal comfort has a significant influence on a building occupant's overall well-being, productivity, and satisfaction. Due to its subjectivity, thermal comfort cannot be achieved with common building control strategies, such as defining temperature set-points or averaging across all occupants. Personal control models allow occupants to influence their task environments based on their own preferences and the use of machine learning methods. In this paper, we present LATEST, a system that collects thermal comfort-related data, preprocesses it, and generates a personal temperature control model that controls occupant-specific thermal actuators. We conducted an empirical field study with three human subjects operating task heaters over a period of six weeks to collect data and generate personal thermal control models. These thermal control models were then used to compare the performance of the models with the manual actuation of the task heaters. The evaluation showed that LATEST can reduce occupant command frequency by 79%, while increasing thermal comfort by 9%, compared with manual control. We make the data set and source code available to the public.